In allusion to phenomenon of large instability of underground goaf, research is carried on prediction of instability and mechanism of multi-factor comprehensive action. Firstly, according to large number of examples of instability disaster in mines at home and abroad, influence factors of goaf stability was analyzed, including characteristics of surrounding rock, conditions of engineering geology, engineering properties, structure parameters of goaf and other artificial mining and time etc; then, using BP neural network to establish the comprehensive function matrix RSE, and putting forward a prediction index Fi, this index can be real-time tracking on the actual change conditions of surrounding rock scene in mining goaf, showing the comprehensive action results of each factor on steady state of underground goaf engineering system; finally, on the basis of 36 samples, the stable value and interval value of the corresponding multi-factor comprehensive action index Fi can be obtained after calculation by using comprehensive function matrix RSE. An actual example showed that, value of multi-factor comprehensive action index Fi was-0.8159 in goaf of an iron mine, which was in instability state.